Revert: Raytracing: change intel_get_implicit_dispatch_globals function to use subdevice (aka Tile) id

On multi-tile GPUs with implicit scaling enabled, when we use implicit rtDispatchGlobals pointer the runtime will send an array of pointers, one per tile. This change makes intel_get_implicit_dispatch_globals return the correct pointer for current tile.

Author: PawelJurek

IGC/AdaptorOCL/ocl_igc_shared/raytracing/constants.h

@@ -449,7 +449,6 @@ set(IGC_BUILD__BIF_OCL_COMMON_INC_DIRS
     "${IGC_OPTION__BIF_SRC_OCL_DIR}/Languages/OpenCL"
     "${IGC_OPTION__BIF_SRC_OCL_DIR}/Languages/OpenCL/PointerSize"
     "${IGC_OPTION__BIF_SRC_OCL_DIR}/Headers"
-    "${CMAKE_CURRENT_SOURCE_DIR}/../AdaptorOCL/ocl_igc_shared"
     "${CMAKE_CURRENT_SOURCE_DIR}/../AdaptorOCL/ocl_igc_shared/device_enqueue"
   )
 

IGC/BiFModule/CMakeLists.txt

@@ -5485,8 +5485,5 @@ void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(prefetch, _p1v16f64_i64, )( global dou
 uint  __builtin_spirv_OpReadClockKHR_i32_i32(uint scope);
 ulong __builtin_spirv_OpReadClockKHR_i64_i32(uint scope);
 
-int SPIRV_OVERLOADABLE SPIRV_BUILTIN_NO_OP(BuiltInSubDeviceIDINTEL, , )(void);
-int SPIRV_OVERLOADABLE SPIRV_BUILTIN_NO_OP(GlobalHWThreadIDINTEL, , )(void);
-
 #endif // __SPIRV_H__
 

IGC/BiFModule/Headers/spirv.h

@@ -868,7 +868,7 @@ void* __builtin_IB_intel_get_global_btd_stack(rtglobals_t rt_dispatch_globals);
 rtfence_t __builtin_IB_intel_dispatch_trace_ray_query(
   rtglobals_t rt_dispatch_globals, uint bvh_level, uint traceTayCtrl);
 void __builtin_IB_intel_rt_sync(rtfence_t fence);
-global void* __builtin_IB_intel_get_rt_global_buffer();
+global void* __builtin_IB_intel_get_implicit_dispatch_globals();
 #endif // cl_intel_pvc_rt_validation
 
 void    __builtin_IB_hdc_uncompressed_write_uchar(__global uchar *buf, uchar val);

IGC/BiFModule/Implementation/IGCBiF_Intrinsics.cl

@@ -8,8 +8,6 @@ SPDX-License-Identifier: MIT
 
 #ifdef cl_intel_pvc_rt_validation
 
-#include "raytracing/constants.h"
-
 void* intel_get_rt_stack( rtglobals_t rt_dispatch_globals ) {
   return __builtin_IB_intel_get_rt_stack(rt_dispatch_globals);
 }
@@ -32,9 +30,7 @@ void intel_rt_sync(rtfence_t fence) {
 }
 
 global void* intel_get_implicit_dispatch_globals() {
-    global char* globalBuffer = __builtin_IB_intel_get_rt_global_buffer();
-    int subDeviceID = SPIRV_BUILTIN_NO_OP(BuiltInSubDeviceIDINTEL, , )();
-    return globalBuffer + subDeviceID * DISPATCH_GLOBALS_STRIDE;
+    return __builtin_IB_intel_get_implicit_dispatch_globals();
 }
 
 #endif

IGC/BiFModule/Languages/OpenCL/PreRelease/IBiF_raytracing.cl

@@ -38,7 +38,7 @@ namespace {
     {"__builtin_IB_intel_get_global_btd_stack",          &ResolveOCLRaytracingBuiltins::handleGetGlobalBTDStack     },
     {"__builtin_IB_intel_dispatch_trace_ray_query",      &ResolveOCLRaytracingBuiltins::handleDispatchTraceRayQuery },
     {"__builtin_IB_intel_rt_sync",                       &ResolveOCLRaytracingBuiltins::handleRTSync                },
-    {"__builtin_IB_intel_get_rt_global_buffer",          &ResolveOCLRaytracingBuiltins::handleGetRTGlobalBuffer     },
+    {"__builtin_IB_intel_get_implicit_dispatch_globals", &ResolveOCLRaytracingBuiltins::handleGetImplicitDG         },
   };
 }
 
@@ -332,12 +332,12 @@ void ResolveOCLRaytracingBuiltins::handleRTSync(CallInst& callInst) {
 
 /*
 Handler for
-void __builtin_IB_intel_get_rt_global_buffer();
+void __builtin_IB_intel_get_implicit_dispatch_globals();
 
 Description:
 Returns IMPLICIT_RT_GLOBAL_BUFFER implicit argument.
 */
-void ResolveOCLRaytracingBuiltins::handleGetRTGlobalBuffer(llvm::CallInst& callInst) {
+void ResolveOCLRaytracingBuiltins::handleGetImplicitDG(llvm::CallInst& callInst) {
     RTBuilder rtbuilder(m_builder->getContext(), *m_pCtx);
     rtbuilder.SetInsertPoint(&callInst);
     auto v = rtbuilder.getGlobalBufferPtr();

IGC/Compiler/Optimizer/OpenCLPasses/RayTracing/ResolveOCLRaytracingBuiltins.cpp

@@ -49,7 +49,7 @@ namespace IGC {
     void handleGetGlobalBTDStack(llvm::CallInst& callInst);
     void handleDispatchTraceRayQuery(llvm::CallInst& callInst);
     void handleRTSync(llvm::CallInst& callInst);
-    void handleGetRTGlobalBuffer(llvm::CallInst& callInst);
+    void handleGetImplicitDG(llvm::CallInst& callInst);
 
   private:
     CodeGenContext* m_pCtx;